A data-driven characterization of municipal water uses in the contiguous United States of America
Date
2024
Journal Title
Journal ISSN
Volume Title
Abstract
Municipal water systems in the United States (U.S.) are facing increasing challenges due to changing urban population dynamics and socio-economic conditions as well as from the impacts of weather extremities on water availability and quality. These challenges pose a serious risk to the municipal water providers by hindering their ability to continue providing safe drinking water to residents while also securing adequate supply for economic growth. A data-driven approach has been developed in this study to characterize the trends, patterns, and urban scaling relationships in municipal water consumption across the Contiguous United States. Then using sophisticated and robust statistical methods, water consumption patterns are modeled, identifying key climatic, socio-economic, and regional factors. The first chapter of this data-driven study looked at municipal water uses of 126 cities and towns across the U.S. from 2005 to 2017, analyzing the temporal trends and spatial patterns in water consumption and identifying the influencing factors. Water usage in gallons per person per day, ratio of commercial, industrial, and institutional (CII) to Residential water use, and percent outdoor water consumption were statistically calculated using aggregated monthly and annual water use data. The end goal was to statistically relate the variations in CII to Residential water use ratio across the municipalities with their local climatic, socio-economic, and regional factors. The results indicate an overall decreasing trend in municipal water use, 2.6 gallons per person annually, with greater reductions achieved in the residential sector. Both Residential and CII water use exhibit significant seasonality over an average year. Large cities, particularly in the southern and western parts of the U.S. with arid climates, had the highest demand for water but also showed the largest annual reductions in their per capita water consumption. This study also revealed that outdoor water use varied significantly from 3 to 64 percent of the Total water consumption across the U.S., and it was highest in smaller cities in the western and arid regions. Factors such as April precipitation, annual vapor pressure deficit, number of employees in the manufacturing sector, total percentage of houses built before 1950, and total percentage of single-family houses explain much of the variation in CII to Residential water use ratio across the CONUS. The second chapter leverages high-resolution, smart-metered water use data from over 900 single-family households in Arizona for the water year 2021. This part of the study characterizes the determinants or drivers of water consumption patterns, specifically in single-family households, and presents a framework of statistical methods for analyzing smart-metered water consumption data in future research. A novel approach was developed to characterize household appliance efficiency levels using clustering techniques on 5-second interval data. Integrating water consumption data with detailed spatial information of the household and building characteristics, along with local climatic factors, yielded a robust mixed-effects model that captured the variations in household water uses with high accuracy at a monthly time-step. Local air temperature, household occupancy level, presence of a swimming pool, the year the household was built, and the efficiency of indoor appliances and irrigation systems were exhibited to be the key factors influencing variations in household water use. The third and fourth chapter of this study reanalyzed the water consumption data of those 126 municipalities. The third chapter dwelled into the estimation of the state of water consumption efficiencies or economics of scale in the municipal water systems using an econometrics framework called urban scaling theory. A parsimonious mixed-effects model that combined the effects of socio-economic, built environment, and regional factors, such as climate zones and water use type, was developed to model annual water uses. The results confirm efficiencies in water systems as cities grow and become denser, with CII water use category showing the highest efficiency gains followed by the Residential and Total water use categories. A key finding is the estimation of the unique variations in water use efficiency patterns across the U.S. These variations are influenced by factors such as population, housing characteristics, the combined effects of climate type and geographical location of the cities, and the type of water use category (Residential or CII) that dominates in each city. The fourth or the final chapter synthesizes the lessons learned previously about the drivers of municipal water uses and explores the development of a model for predicting monthly water consumption patterns using machine learning algorithms. These algorithms demonstrated improved capabilities in predicting the Total monthly water use more accurately than the previous modeling efforts, also controlling for factors with multi-collinearity. Climatic variables (like precipitation and vapor pressure deficit), socio-economic and built environment variables (such as income level and housing characteristics), and regional factors (including climate type and water use type dominance in a city), were confirmed by the machine learning algorithms to strongly influence and cause variations in the municipal water consumption patterns. Overall, this study showcases the power of data-driven approaches to effectively understand the nuances in municipal water uses. Integration of the lessons learned and the statistical frameworks used in this study can empower water utilities and city planners to manage municipal water demands with greater resiliency and efficiency.
Description
Rights Access
Embargo expires: 05/20/2025.
Subject
machine learning
municipal water consumption
water demand management
mixed-effects modeling
CII water use
residenital water use